From intent to execution
SAP's TechEd 2026 keynote centered on Joule Work, an AI-native work environment designed to let employees state what they want in natural language while agents handle the individual steps across SAP and non-SAP systems. It is available through browser, desktop, mobile, voice, and even Apple CarPlay, and SAP is moving it toward general availability now rather than treating it as an early-stage preview. The company paired it with Joule Spaces, AI-generated workspaces tied to specific business outcomes that refresh automatically with current data and can be shared across a team, extending the same intent-driven model to collaborative work rather than individual tasks.
SAP chief technology officer Philipp Herzig summarized the strategic shift in blunt terms: this year, we are showing how we are moving from AI learning your business to running it. Atul Deo, SVP and global head of AI product management and partnerships, added useful context on why that shift matters operationally, noting that users previously had to understand their specific applications and user experiences just to get work done, and that the new model inverts that requirement by having agents navigate the applications on the user's behalf. Deo also argued that context, not model choice, is now the real differentiator between competing AI platforms.
The adoption numbers behind the keynote
SAP reported about 200 customers currently in early adopter care for Joule Work, a relatively modest figure against SAP's global customer base of well over 20,000 enterprises, and more than 70,000 SAP employees using the platform internally. The keynote also cited 40 assistants and 300 agents already released as part of the broader Autonomous Enterprise initiative, a number that sounds impressive until you ask how many of those 300 agents are running in actual production workflows versus sitting in a partner catalog waiting for a customer to activate them. SAP did not break that figure down by deployment status.
The gap between keynote scale and verified production adoption is the detail enterprise buyers should press hardest in any SAP conversation this quarter. Internal employee usage at 70,000 is a meaningful signal, since SAP is effectively running its own operations on the platform it is selling, which is a stronger form of validation than a vendor simply claiming capability. But internal dogfooding and external paying customer adoption are different tests, and the 200 early adopter figure is the number that actually indicates how many organizations have moved past evaluation into committed deployment.
Governance built for a specific regulation
SAP paired the agentic capability announcements with a governance stack that includes an SAP Enterprise AI Gateway, enhanced controls inside the SAP AI Agent Hub, and an AI Governance Assistant explicitly built around EU AI Act requirements. That is a notably more specific regulatory target than most vendors are naming this year, and it reflects SAP's large European customer base, where the AI Act's compliance deadlines carry direct legal weight rather than being a general best-practice consideration. Joule Work's agentic execution also runs on open standards, including A2A and MCP, which lets third-party agents invoke SAP's agents through what SAP describes as a governed framework.
For enterprises operating under EU AI Act obligations, SAP naming that regulation directly in its governance tooling is a meaningful signal worth following up on with specifics. Vendors frequently describe governance features in generic terms, leaving the customer to work out independently whether the tooling actually satisfies a particular jurisdiction's requirements. CIOs and compliance teams evaluating SAP's agentic rollout should ask for documentation mapping the AI Governance Assistant's controls directly to specific EU AI Act articles, rather than accepting the regulatory framing at face value during a sales conversation.
Betting on an ecosystem of forward-deployed partners
SAP highlighted partner integrations from Intel, Docusign, Parloa, and Redwood during the keynote, and the company's broader strategy depends heavily on this kind of partner ecosystem delivering industry-specific agentic workflows that SAP itself has no incentive to build directly. That is a deliberate division of labor: SAP builds the governed platform, the open standards for agent communication, and the underlying data layer, while partners build the vertical-specific automation that actually shows up in a customer's day-to-day operations. SAP also pointed to an unusual proof point for its forecasting model, SAP-RPT-TS, deployed with HPE on the International Space Station for mission-health monitoring, a demonstration clearly meant to signal technical credibility rather than near-term relevance to most enterprise customers.
The partner-dependent model carries its own risk for enterprise buyers. A customer's actual business value depends heavily on whether a specific partner has already built the workflow that company needs, even when the underlying governance and data layer is strong. Enterprises evaluating SAP's Autonomous Enterprise push should ask about which partners are building in their specific industry and how mature those partner-built agents actually are today, since the keynote's 300 agents figure almost certainly spans a wide range of maturity levels across that partner ecosystem.
The developer and data layer underneath
Beneath the headline agentic features, SAP announced Joule Studio, an AI-first development environment combining no-code and pro-code tools with support for more than 70 AI models, integration with the n8n workflow automation tool, and new observability and cost controls for monitoring agent spend. On the data side, SAP expanded its Knowledge Graph to cover customer-specific data, introduced an Ontology Manager for its Business Data Cloud, deepened lakehouse integration with Dremio, broadened its Databricks partnership, and added SAP Reltio for multidomain master data management. SAP also introduced two new foundation models, SAP-RPT-2.0 for anomaly detection and structured data analysis and SAP-RPT-TS for time-series forecasting.
The data layer announcements are arguably more consequential long-term than the headline agent features, because agentic AI is only as good as the business context it can draw on, and SAP's own executives made that argument explicitly during the keynote. Atul Deo's point that context matters more than model choice is a direct answer to competitors emphasizing raw model capability, and it reframes the competitive question for enterprise buyers: the relevant comparison between ERP vendors' AI strategies may have less to do with which large language model each one uses and more to do with how cleanly each vendor's knowledge graph and master data actually represent a customer's real business.
